Things I've built
A few projects that show what I actually build.
Two AI platforms, an open-source TypeScript ORM, and a Drupal module on 99,000+ sites. Click through and poke at any of them.
Variability.ai
Records meetings, produces transcripts and summaries in 45+ languages. Ask questions like "what did we decide about X?"
More details
Variability records your meetings and turns them into something useful: a transcript, an instant summary, the action items, and a chat you can ask "what did we actually decide about X?". It hooks into Google Meet, Zoom, and Teams, and also chews through anything you import, from a YouTube link to a raw MP4. I built it end to end — the pipeline that ingests and transcribes the media, the chat layer on top, and the product around them. It works across 45+ languages and has a summary ready within about 30 seconds of you hitting stop.
Infrahub.ai
Scores road conditions from smartphone photos. Turns thousands of raw images into actionable maintenance plans for cities.
More details
Infrahub lets transportation agencies assess road conditions from smartphone photos instead of sending crews out with clipboards. Field staff snap pictures, the AI scores the surface, and the platform ranks what to fix first against the available budget. I worked on the data platform behind it: the part that turns thousands of raw images into a maintenance plan a city can actually act on.
UQL ORM
Open-source query protocol for TypeScript. Queries are plain JSON — no DSL, no code generation. 3.9M ops/sec, 2.4x faster than the next ORM.
More details
UQL is my open-source query protocol and ORM for TypeScript. The whole thing rests on one idea: a query is just a plain JSON object, so there is no DSL to learn and no code-generation step. You write a query once and run it anywhere, from Node, Bun, and Deno to Cloudflare Workers, React Native, and the browser, against Postgres, MySQL, MariaDB, SQLite, LibSQL, Neon, D1, and MongoDB. It is type-safe end to end and ships native vector and semantic search for AI work. In my own benchmarks it does roughly 3.9M ops/sec, about 2.4x the next ORM. I started it because every project I joined had quietly reinvented the same fragile data layer.
EasyBreadcrumb
Drupal module that automates breadcrumb trails since 2012. JSON-LD for SEO, taxonomy hierarchies. Running on 99,000+ live sites.
More details
EasyBreadcrumb is a Drupal module I created back in 2012. It builds breadcrumb trails automatically from the page URL and title, with JSON-LD structured data baked in for SEO, taxonomy-aware hierarchies, and per-segment tweaks like truncation and regex replacement. Site builders install it and stop hand-wiring breadcrumbs. It outgrew me a long time ago: it currently runs on roughly 99,000 live Drupal sites across government, education, and enterprise, and supports Drupal 9 through 11.